Face Generation¶

In this project, you'll use generative adversarial networks to generate new images of faces.

Get the Data¶

You'll be using two datasets in this project:

MNIST
CelebA

Since the celebA dataset is complex and you're doing GANs in a project for the first time, we want you to test your neural network on MNIST before CelebA. Running the GANs on MNIST will allow you to see how well your model trains sooner.

If you're using FloydHub, set data_dir to "/input" and use the FloydHub data ID "R5KrjnANiKVhLWAkpXhNBe".

data_dir = './data'

# FloydHub - Use with data ID "R5KrjnANiKVhLWAkpXhNBe"
data_dir = '/input'


"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import helper

helper.download_extract('mnist', data_dir)
helper.download_extract('celeba', data_dir)

Found mnist Data
Found celeba Data

Explore the Data¶

MNIST¶

As you're aware, the MNIST dataset contains images of handwritten digits. You can view the first number of examples by changing show_n_images.

show_n_images = 25

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
%matplotlib inline
import os
from glob import glob
from matplotlib import pyplot

mnist_images = helper.get_batch(glob(os.path.join(data_dir, 'mnist/*.jpg'))[:show_n_images], 28, 28, 'L')
pyplot.imshow(helper.images_square_grid(mnist_images, 'L'), cmap='gray')

<matplotlib.image.AxesImage at 0x7f171c898278>

CelebA¶

The CelebFaces Attributes Dataset (CelebA) dataset contains over 200,000 celebrity images with annotations. Since you're going to be generating faces, you won't need the annotations. You can view the first number of examples by changing show_n_images.

show_n_images = 25

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
mnist_images = helper.get_batch(glob(os.path.join(data_dir, 'img_align_celeba/*.jpg'))[:show_n_images], 28, 28, 'RGB')
pyplot.imshow(helper.images_square_grid(mnist_images, 'RGB'))

<matplotlib.image.AxesImage at 0x7f178fce82e8>

Preprocess the Data¶

Since the project's main focus is on building the GANs, we'll preprocess the data for you. The values of the MNIST and CelebA dataset will be in the range of -0.5 to 0.5 of 28x28 dimensional images. The CelebA images will be cropped to remove parts of the image that don't include a face, then resized down to 28x28.

The MNIST images are black and white images with a single color channel while the CelebA images have 3 color channels (RGB color channel).

Build the Neural Network¶

You'll build the components necessary to build a GANs by implementing the following functions below:

model_inputs
discriminator
generator
model_loss
model_opt
train

Check the Version of TensorFlow and Access to GPU¶

This will check to make sure you have the correct version of TensorFlow and access to a GPU

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
from distutils.version import LooseVersion
import warnings
import tensorflow as tf

# Check TensorFlow Version
assert LooseVersion(tf.__version__) >= LooseVersion('1.0'), 'Please use TensorFlow version 1.0 or newer.  You are using {}'.format(tf.__version__)
print('TensorFlow Version: {}'.format(tf.__version__))

# Check for a GPU
if not tf.test.gpu_device_name():
    warnings.warn('No GPU found. Please use a GPU to train your neural network.')
else:
    print('Default GPU Device: {}'.format(tf.test.gpu_device_name()))

TensorFlow Version: 1.1.0
Default GPU Device: /gpu:0

Input¶

Implement the model_inputs function to create TF Placeholders for the Neural Network. It should create the following placeholders:

Real input images placeholder with rank 4 using image_width, image_height, and image_channels.
Z input placeholder with rank 2 using z_dim.
Learning rate placeholder with rank 0.

Return the placeholders in the following the tuple (tensor of real input images, tensor of z data)

import problem_unittests as tests

def model_inputs(image_width, image_height, image_channels, z_dim):
    """
    Create the model inputs
    :param image_width: The input image width
    :param image_height: The input image height
    :param image_channels: The number of image channels
    :param z_dim: The dimension of Z
    :return: Tuple of (tensor of real input images, tensor of z data, learning rate)
    """
    inputs_real = tf.placeholder(tf.float32, (None, image_height, image_width, image_channels), name='real_inputs')
    inputs_z = tf.placeholder(tf.float32, (None, z_dim), name='z_inputs')
    learn_rate = tf.placeholder(tf.float32, name='learning_rate')
        
    return inputs_real, inputs_z, learn_rate



"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_inputs(model_inputs)

Tests Passed

Discriminator¶

Implement discriminator to create a discriminator neural network that discriminates on images. This function should be able to reuse the variables in the neural network. Use tf.variable_scope with a scope name of "discriminator" to allow the variables to be reused. The function should return a tuple of (tensor output of the discriminator, tensor logits of the discriminator).

def discriminator(images, reuse=False):
    """
    Create the discriminator network
    :param images: Tensor of input image(s)
    :param reuse: Boolean if the weights should be reused
    :return: Tuple of (tensor output of the discriminator, tensor logits of the discriminator)
    """
    strides = 2
    with tf.variable_scope('discriminator', reuse=reuse):
        # Input layer is 28x28x3
        layer1 = tf.layers.conv2d(images, 64, 5, strides=strides, padding='same')
        relu1 = tf.maximum(0.18 * layer1, layer1)
        # 14x14x64
        
        layer2 = tf.layers.conv2d(relu1, 128, 5, strides=strides, padding='same')
        normalised2 = tf.layers.batch_normalization(layer2, training=True)
        relu2 = tf.maximum(0.18 * normalised2, normalised2)
        # 7x7x128
        
        layer3 = tf.layers.conv2d(relu2, 256, 5, strides=strides, padding='same')
        normalised3 = tf.layers.batch_normalization(layer3, training=True)
        relu3 = tf.maximum(0.18 * normalised3, normalised3)
        # 4x4x256

        # Flatten it
        flat = tf.reshape(relu3, (-1, 4*4*256))
        logits = tf.layers.dense(flat, 1)
        out = tf.sigmoid(logits)
        
        return out, logits
    
    

"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_discriminator(discriminator, tf)

Tests Passed

Generator¶

Implement generator to generate an image using z. This function should be able to reuse the variables in the neural network. Use tf.variable_scope with a scope name of "generator" to allow the variables to be reused. The function should return the generated 28 x 28 x out_channel_dim images.

def generator(z, out_channel_dim, is_train=True):
    """
    Create the generator network
    :param z: Input z
    :param out_channel_dim: The number of channels in the output image
    :param is_train: Boolean if generator is being used for training
    :return: The tensor output of the generator
    """
    with tf.variable_scope('generator', reuse=not is_train):

        # -> 4x4x512
        layer1 = tf.layers.dense(z, 4*4*512)
        layer1 = tf.reshape(layer1, (-1, 4, 4, 512))
        normalised1 = tf.layers.batch_normalization(layer1, training=is_train)
        leaky_relu1 = tf.maximum(0.18 * normalised1, normalised1)
        
        # -> 7x7x256
        layer2 = tf.image.resize_nearest_neighbor(leaky_relu1, (7,7))
        convolution2 = tf.layers.conv2d(layer2, 256, 5, padding='same', activation=None)
        normalised2 = tf.layers.batch_normalization(convolution2, training=is_train)
        leaky_relu2 = tf.maximum(0.18 * normalised2, normalised2)
        
        # -> 16x16x128
        layer3 = tf.image.resize_nearest_neighbor(leaky_relu2, (14,14))
        conv3 = tf.layers.conv2d(layer3, 128, 5, padding='same', activation=None)
        normalised3 = tf.layers.batch_normalization(conv3, training=is_train)
        leaky_relu3 = tf.maximum(0.18 * normalised3, normalised3)
        
        # -> 32x32x3
        layer4 = tf.image.resize_nearest_neighbor(leaky_relu3, (28,28))
        logits = tf.layers.conv2d(layer4, out_channel_dim, 5, padding='same')
        
        return tf.tanh(logits)
        
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_generator(generator, tf)

Tests Passed

Loss¶

Implement model_loss to build the GANs for training and calculate the loss. The function should return a tuple of (discriminator loss, generator loss). Use the following functions you implemented:

discriminator(images, reuse=False)
generator(z, out_channel_dim, is_train=True)

def model_loss(input_real, input_z, out_channel_dim):
    """
    Get the loss for the discriminator and generator
    :param input_real: Images from the real dataset
    :param input_z: Z input
    :param out_channel_dim: The number of channels in the output image
    :return: A tuple of (discriminator loss, generator loss)
    """    
    # Calculate losses
    generator_model = generator(input_z, out_channel_dim, is_train=True)
    discriminator_model_real, discriminator_logits_real = discriminator(input_real, reuse=False)
    # Fake output generated by the generator model
    discriminator_model_fake, discriminator_logits_fake = discriminator(generator_model, reuse=True)

    # Discriminator loss
    discriminator_loss_real = tf.reduce_mean(
        tf.nn.sigmoid_cross_entropy_with_logits(logits=discriminator_logits_real, labels = tf.ones_like(discriminator_model_real) * 0.9))
    discriminator_loss_fake = tf.reduce_mean(
        tf.nn.sigmoid_cross_entropy_with_logits(logits=discriminator_logits_fake, labels = tf.zeros_like(discriminator_model_fake)))
    discriminator_loss = discriminator_loss_fake + discriminator_loss_real
    
    # Generator loss
    generator_loss = tf.reduce_mean(
        tf.nn.sigmoid_cross_entropy_with_logits(logits=discriminator_logits_fake, labels = tf.ones_like(discriminator_model_fake)))

    return discriminator_loss, generator_loss



"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_loss(model_loss)

Tests Passed

Optimization¶

Implement model_opt to create the optimization operations for the GANs. Use tf.trainable_variables to get all the trainable variables. Filter the variables with names that are in the discriminator and generator scope names. The function should return a tuple of (discriminator training operation, generator training operation).

def model_opt(d_loss, g_loss, learning_rate, beta1):
    """
    Get optimization operations
    :param d_loss: Discriminator loss Tensor
    :param g_loss: Generator loss Tensor
    :param learning_rate: Learning Rate Placeholder
    :param beta1: The exponential decay rate for the 1st moment in the optimizer
    :return: A tuple of (discriminator training operation, generator training operation)
    """
    # Get the trainable_variables, split into Generator and Discriminator parts
    t_vars = tf.trainable_variables()
    g_vars = [var for var in t_vars if var.name.startswith('generator')]
    d_vars = [var for var in t_vars if var.name.startswith('discriminator')]

    d_train_opt = tf.train.AdamOptimizer(learning_rate, beta1=beta1).minimize(d_loss, var_list=d_vars)
    g_train_opt = tf.train.AdamOptimizer(learning_rate, beta1=beta1).minimize(g_loss, var_list=g_vars)
    
    return d_train_opt, g_train_opt


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_opt(model_opt, tf)

Tests Passed

Neural Network Training¶

Show Output¶

Use this function to show the current output of the generator during training. It will help you determine how well the GANs is training.

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import numpy as np

def show_generator_output(sess, n_images, input_z, out_channel_dim, image_mode):
    """
    Show example output for the generator
    :param sess: TensorFlow session
    :param n_images: Number of Images to display
    :param input_z: Input Z Tensor
    :param out_channel_dim: The number of channels in the output image
    :param image_mode: The mode to use for images ("RGB" or "L")
    """
    cmap = None if image_mode == 'RGB' else 'gray'
    z_dim = input_z.get_shape().as_list()[-1]
    example_z = np.random.uniform(-1, 1, size=[n_images, z_dim])

    samples = sess.run(
        generator(input_z, out_channel_dim, False),
        feed_dict={input_z: example_z})

    images_grid = helper.images_square_grid(samples, image_mode)
    pyplot.imshow(images_grid, cmap=cmap)
    pyplot.show()

Train¶

Implement train to build and train the GANs. Use the following functions you implemented:

model_inputs(image_width, image_height, image_channels, z_dim)
model_loss(input_real, input_z, out_channel_dim)
model_opt(d_loss, g_loss, learning_rate, beta1)

Use the show_generator_output to show generator output while you train. Running show_generator_output for every batch will drastically increase training time and increase the size of the notebook. It's recommended to print the generator output every 100 batches.

def train(epoch_count, batch_size, z_dim, learning_rate, beta1, get_batches, data_shape, data_image_mode):
    """
    Train the GAN
    :param epoch_count: Number of epochs
    :param batch_size: Batch Size
    :param z_dim: Z dimension
    :param learning_rate: Learning Rate
    :param beta1: The exponential decay rate for the 1st moment in the optimizer
    :param get_batches: Function to get batches
    :param data_shape: Shape of the data
    :param data_image_mode: The image mode to use for images ("RGB" or "L")
    """    
    losses = []
    steps = 0
    n_images = 25
        
    _, image_width, image_height, image_channels = data_shape
    input_real, input_z , learning_rate_pl = model_inputs(image_width,image_height,image_channels,z_dim)
    
    d_loss, g_loss = model_loss(input_real, input_z, image_channels)
    d_opt, g_opt = model_opt(d_loss, g_loss,learning_rate_pl, beta1)
    
    with tf.Session() as sess:
        sess.run(tf.global_variables_initializer())
        for epoch_i in range(epoch_count): 
            for batch_images in get_batches(batch_size):
                steps = steps + 1
                
                batch_z = np.random.uniform(-1, 1, size=(batch_size, z_dim))
                batch_images = batch_images * 2.0
                
                _ = sess.run(d_opt, feed_dict={input_real:batch_images,input_z:batch_z,learning_rate_pl:learning_rate})
                _ = sess.run(g_opt, feed_dict={input_real:batch_images,input_z:batch_z,learning_rate_pl:learning_rate})
                
                if steps % 5 == 0 :
                    train_loss_d = d_loss.eval({input_real:batch_images,input_z:batch_z,learning_rate_pl:learning_rate})
                    train_loss_g = g_loss.eval({input_real:batch_images,input_z:batch_z,learning_rate_pl:learning_rate})
                    print("Epoch {}/{} ___ ".format(epoch_i+1, epoch_count),
                      "Discriminator Loss: {:.4f} ___".format(train_loss_d),
                      "Generator Loss: {:.4f}".format(train_loss_g))
                
                    losses.append((train_loss_d, train_loss_g))
                    
                if steps % 50 == 0:
                    show_generator_output(sess,n_images,input_z,image_channels,data_image_mode)

MNIST¶

Test your GANs architecture on MNIST. After 2 epochs, the GANs should be able to generate images that look like handwritten digits. Make sure the loss of the generator is lower than the loss of the discriminator or close to 0.

batch_size = 16
z_dim = 128 
learning_rate = 0.0002 
beta1 = 0.4


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 2

mnist_dataset = helper.Dataset('mnist', glob(os.path.join(data_dir, 'mnist/*.jpg')))
with tf.Graph().as_default():
    train(epochs, batch_size, z_dim, learning_rate, beta1, mnist_dataset.get_batches,
          mnist_dataset.shape, mnist_dataset.image_mode)

Epoch 1/2 ___  Discriminator Loss: 0.8352 ___ Generator Loss: 1.3306
Epoch 1/2 ___  Discriminator Loss: 0.9981 ___ Generator Loss: 1.4084
Epoch 1/2 ___  Discriminator Loss: 1.5683 ___ Generator Loss: 1.5027
Epoch 1/2 ___  Discriminator Loss: 0.9054 ___ Generator Loss: 1.5043
Epoch 1/2 ___  Discriminator Loss: 1.0909 ___ Generator Loss: 1.9228
Epoch 1/2 ___  Discriminator Loss: 1.1001 ___ Generator Loss: 1.6476
Epoch 1/2 ___  Discriminator Loss: 2.0118 ___ Generator Loss: 1.6230
Epoch 1/2 ___  Discriminator Loss: 2.1759 ___ Generator Loss: 0.2521
Epoch 1/2 ___  Discriminator Loss: 1.7394 ___ Generator Loss: 0.4467
Epoch 1/2 ___  Discriminator Loss: 1.8242 ___ Generator Loss: 0.4159

Epoch 1/2 ___  Discriminator Loss: 2.9802 ___ Generator Loss: 0.1272
Epoch 1/2 ___  Discriminator Loss: 1.5471 ___ Generator Loss: 0.5927
Epoch 1/2 ___  Discriminator Loss: 2.2794 ___ Generator Loss: 2.9082
Epoch 1/2 ___  Discriminator Loss: 2.4898 ___ Generator Loss: 0.2813
Epoch 1/2 ___  Discriminator Loss: 1.6485 ___ Generator Loss: 0.6405
Epoch 1/2 ___  Discriminator Loss: 2.8639 ___ Generator Loss: 0.1161
Epoch 1/2 ___  Discriminator Loss: 2.2110 ___ Generator Loss: 0.2178
Epoch 1/2 ___  Discriminator Loss: 1.8860 ___ Generator Loss: 0.3231
Epoch 1/2 ___  Discriminator Loss: 1.4933 ___ Generator Loss: 0.6580
Epoch 1/2 ___  Discriminator Loss: 1.5233 ___ Generator Loss: 0.7904

Epoch 1/2 ___  Discriminator Loss: 1.5934 ___ Generator Loss: 0.3930
Epoch 1/2 ___  Discriminator Loss: 1.2163 ___ Generator Loss: 0.6205
Epoch 1/2 ___  Discriminator Loss: 1.4650 ___ Generator Loss: 1.0694
Epoch 1/2 ___  Discriminator Loss: 1.8000 ___ Generator Loss: 0.4010
Epoch 1/2 ___  Discriminator Loss: 1.7267 ___ Generator Loss: 0.4187
Epoch 1/2 ___  Discriminator Loss: 1.4546 ___ Generator Loss: 0.6208
Epoch 1/2 ___  Discriminator Loss: 1.7171 ___ Generator Loss: 0.3290
Epoch 1/2 ___  Discriminator Loss: 1.4464 ___ Generator Loss: 0.5036
Epoch 1/2 ___  Discriminator Loss: 1.4046 ___ Generator Loss: 0.5636
Epoch 1/2 ___  Discriminator Loss: 1.6475 ___ Generator Loss: 0.5746

Epoch 1/2 ___  Discriminator Loss: 1.3102 ___ Generator Loss: 0.6506
Epoch 1/2 ___  Discriminator Loss: 1.5143 ___ Generator Loss: 0.4898
Epoch 1/2 ___  Discriminator Loss: 1.4359 ___ Generator Loss: 0.6453
Epoch 1/2 ___  Discriminator Loss: 1.7737 ___ Generator Loss: 0.3367
Epoch 1/2 ___  Discriminator Loss: 1.3132 ___ Generator Loss: 1.4067
Epoch 1/2 ___  Discriminator Loss: 1.6263 ___ Generator Loss: 0.4639
Epoch 1/2 ___  Discriminator Loss: 1.7040 ___ Generator Loss: 0.3268
Epoch 1/2 ___  Discriminator Loss: 1.8786 ___ Generator Loss: 0.3766
Epoch 1/2 ___  Discriminator Loss: 1.4499 ___ Generator Loss: 0.7313
Epoch 1/2 ___  Discriminator Loss: 1.6615 ___ Generator Loss: 0.3996

Epoch 1/2 ___  Discriminator Loss: 1.6628 ___ Generator Loss: 0.5698
Epoch 1/2 ___  Discriminator Loss: 1.9204 ___ Generator Loss: 0.2733
Epoch 1/2 ___  Discriminator Loss: 1.2025 ___ Generator Loss: 0.8745
Epoch 1/2 ___  Discriminator Loss: 1.2372 ___ Generator Loss: 0.9205
Epoch 1/2 ___  Discriminator Loss: 1.3609 ___ Generator Loss: 0.6725
Epoch 1/2 ___  Discriminator Loss: 1.8803 ___ Generator Loss: 0.3750
Epoch 1/2 ___  Discriminator Loss: 1.8602 ___ Generator Loss: 0.4565
Epoch 1/2 ___  Discriminator Loss: 1.6211 ___ Generator Loss: 0.5806
Epoch 1/2 ___  Discriminator Loss: 1.1710 ___ Generator Loss: 1.0476
Epoch 1/2 ___  Discriminator Loss: 1.3688 ___ Generator Loss: 0.6085

Epoch 1/2 ___  Discriminator Loss: 1.1627 ___ Generator Loss: 0.8849
Epoch 1/2 ___  Discriminator Loss: 1.2029 ___ Generator Loss: 0.7872
Epoch 1/2 ___  Discriminator Loss: 2.2479 ___ Generator Loss: 0.1827
Epoch 1/2 ___  Discriminator Loss: 1.5453 ___ Generator Loss: 0.4821
Epoch 1/2 ___  Discriminator Loss: 1.4186 ___ Generator Loss: 0.5579
Epoch 1/2 ___  Discriminator Loss: 1.2153 ___ Generator Loss: 0.7030
Epoch 1/2 ___  Discriminator Loss: 1.1713 ___ Generator Loss: 0.9076
Epoch 1/2 ___  Discriminator Loss: 1.5135 ___ Generator Loss: 0.5220
Epoch 1/2 ___  Discriminator Loss: 1.3307 ___ Generator Loss: 1.0540
Epoch 1/2 ___  Discriminator Loss: 1.8338 ___ Generator Loss: 0.3592

Epoch 1/2 ___  Discriminator Loss: 1.6408 ___ Generator Loss: 0.3944
Epoch 1/2 ___  Discriminator Loss: 1.2063 ___ Generator Loss: 1.4831
Epoch 1/2 ___  Discriminator Loss: 1.5938 ___ Generator Loss: 0.4637
Epoch 1/2 ___  Discriminator Loss: 1.4526 ___ Generator Loss: 0.5478
Epoch 1/2 ___  Discriminator Loss: 1.4684 ___ Generator Loss: 0.5697
Epoch 1/2 ___  Discriminator Loss: 1.4056 ___ Generator Loss: 0.7044
Epoch 1/2 ___  Discriminator Loss: 1.9906 ___ Generator Loss: 0.2562
Epoch 1/2 ___  Discriminator Loss: 1.5955 ___ Generator Loss: 0.4261
Epoch 1/2 ___  Discriminator Loss: 1.0782 ___ Generator Loss: 1.4271
Epoch 1/2 ___  Discriminator Loss: 1.1365 ___ Generator Loss: 0.8599

Epoch 1/2 ___  Discriminator Loss: 1.6088 ___ Generator Loss: 0.7418
Epoch 1/2 ___  Discriminator Loss: 1.2551 ___ Generator Loss: 0.7584
Epoch 1/2 ___  Discriminator Loss: 1.3764 ___ Generator Loss: 0.5707
Epoch 1/2 ___  Discriminator Loss: 1.6973 ___ Generator Loss: 0.3844
Epoch 1/2 ___  Discriminator Loss: 1.2265 ___ Generator Loss: 0.8098
Epoch 1/2 ___  Discriminator Loss: 1.3599 ___ Generator Loss: 0.7357
Epoch 1/2 ___  Discriminator Loss: 1.6206 ___ Generator Loss: 0.3856
Epoch 1/2 ___  Discriminator Loss: 1.4101 ___ Generator Loss: 0.5694
Epoch 1/2 ___  Discriminator Loss: 1.5856 ___ Generator Loss: 0.4984
Epoch 1/2 ___  Discriminator Loss: 1.1919 ___ Generator Loss: 1.0565

Epoch 1/2 ___  Discriminator Loss: 1.6171 ___ Generator Loss: 0.4146
Epoch 1/2 ___  Discriminator Loss: 1.7871 ___ Generator Loss: 0.3411
Epoch 1/2 ___  Discriminator Loss: 1.0398 ___ Generator Loss: 0.9948
Epoch 1/2 ___  Discriminator Loss: 1.5755 ___ Generator Loss: 0.4806
Epoch 1/2 ___  Discriminator Loss: 0.9972 ___ Generator Loss: 1.1286
Epoch 1/2 ___  Discriminator Loss: 1.2577 ___ Generator Loss: 1.7288
Epoch 1/2 ___  Discriminator Loss: 1.5486 ___ Generator Loss: 0.4882
Epoch 1/2 ___  Discriminator Loss: 1.5315 ___ Generator Loss: 0.6207
Epoch 1/2 ___  Discriminator Loss: 1.3541 ___ Generator Loss: 0.6248
Epoch 1/2 ___  Discriminator Loss: 1.3357 ___ Generator Loss: 0.5850

Epoch 1/2 ___  Discriminator Loss: 1.6894 ___ Generator Loss: 0.3746
Epoch 1/2 ___  Discriminator Loss: 1.4789 ___ Generator Loss: 0.6729
Epoch 1/2 ___  Discriminator Loss: 1.2830 ___ Generator Loss: 0.9742
Epoch 1/2 ___  Discriminator Loss: 1.5933 ___ Generator Loss: 0.4309
Epoch 1/2 ___  Discriminator Loss: 1.2011 ___ Generator Loss: 1.2520
Epoch 1/2 ___  Discriminator Loss: 1.2700 ___ Generator Loss: 1.0125
Epoch 1/2 ___  Discriminator Loss: 1.5331 ___ Generator Loss: 1.0379
Epoch 1/2 ___  Discriminator Loss: 1.2758 ___ Generator Loss: 0.6250
Epoch 1/2 ___  Discriminator Loss: 1.2838 ___ Generator Loss: 0.6671
Epoch 1/2 ___  Discriminator Loss: 1.2379 ___ Generator Loss: 0.7336

CelebA¶

Run your GANs on CelebA. It will take around 20 minutes on the average GPU to run one epoch. You can run the whole epoch or stop when it starts to generate realistic faces.

batch_size = 64
z_dim = 100
learning_rate = 0.0002
beta1 = 0.5


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 1

celeba_dataset = helper.Dataset('celeba', glob(os.path.join(data_dir, 'img_align_celeba/*.jpg')))
with tf.Graph().as_default():
    train(epochs, batch_size, z_dim, learning_rate, beta1, celeba_dataset.get_batches,
          celeba_dataset.shape, celeba_dataset.image_mode)

Epoch 1/1 ___  Discriminator Loss: 1.8460 ___ Generator Loss: 0.4590
Epoch 1/1 ___  Discriminator Loss: 0.6746 ___ Generator Loss: 2.0041
Epoch 1/1 ___  Discriminator Loss: 1.0037 ___ Generator Loss: 2.6666
Epoch 1/1 ___  Discriminator Loss: 0.6782 ___ Generator Loss: 2.1569
Epoch 1/1 ___  Discriminator Loss: 1.0018 ___ Generator Loss: 1.4195
Epoch 1/1 ___  Discriminator Loss: 1.2017 ___ Generator Loss: 0.8457
Epoch 1/1 ___  Discriminator Loss: 0.6880 ___ Generator Loss: 2.3041
Epoch 1/1 ___  Discriminator Loss: 0.4857 ___ Generator Loss: 2.8088
Epoch 1/1 ___  Discriminator Loss: 0.5490 ___ Generator Loss: 2.5631
Epoch 1/1 ___  Discriminator Loss: 0.5782 ___ Generator Loss: 2.4339

Epoch 1/1 ___  Discriminator Loss: 0.5754 ___ Generator Loss: 3.0986
Epoch 1/1 ___  Discriminator Loss: 0.6393 ___ Generator Loss: 2.4090
Epoch 1/1 ___  Discriminator Loss: 1.1990 ___ Generator Loss: 4.9224
Epoch 1/1 ___  Discriminator Loss: 0.8635 ___ Generator Loss: 1.4803
Epoch 1/1 ___  Discriminator Loss: 0.5346 ___ Generator Loss: 2.5229
Epoch 1/1 ___  Discriminator Loss: 2.4752 ___ Generator Loss: 0.1812
Epoch 1/1 ___  Discriminator Loss: 0.8676 ___ Generator Loss: 2.6686
Epoch 1/1 ___  Discriminator Loss: 0.8368 ___ Generator Loss: 1.6427
Epoch 1/1 ___  Discriminator Loss: 0.6257 ___ Generator Loss: 2.1785
Epoch 1/1 ___  Discriminator Loss: 0.7769 ___ Generator Loss: 1.6409

Epoch 1/1 ___  Discriminator Loss: 1.0600 ___ Generator Loss: 1.4315
Epoch 1/1 ___  Discriminator Loss: 1.2017 ___ Generator Loss: 0.8100
Epoch 1/1 ___  Discriminator Loss: 1.1072 ___ Generator Loss: 2.9920
Epoch 1/1 ___  Discriminator Loss: 1.1157 ___ Generator Loss: 2.7165
Epoch 1/1 ___  Discriminator Loss: 1.0353 ___ Generator Loss: 0.9966
Epoch 1/1 ___  Discriminator Loss: 0.7210 ___ Generator Loss: 2.9513
Epoch 1/1 ___  Discriminator Loss: 0.9962 ___ Generator Loss: 1.0990
Epoch 1/1 ___  Discriminator Loss: 1.7712 ___ Generator Loss: 6.4436
Epoch 1/1 ___  Discriminator Loss: 0.9006 ___ Generator Loss: 1.3312
Epoch 1/1 ___  Discriminator Loss: 0.9986 ___ Generator Loss: 1.2579

Epoch 1/1 ___  Discriminator Loss: 1.3507 ___ Generator Loss: 0.6357
Epoch 1/1 ___  Discriminator Loss: 1.2734 ___ Generator Loss: 0.7581
Epoch 1/1 ___  Discriminator Loss: 1.3687 ___ Generator Loss: 0.9410
Epoch 1/1 ___  Discriminator Loss: 1.4080 ___ Generator Loss: 2.9424
Epoch 1/1 ___  Discriminator Loss: 0.7380 ___ Generator Loss: 2.0417
Epoch 1/1 ___  Discriminator Loss: 0.8067 ___ Generator Loss: 1.8391
Epoch 1/1 ___  Discriminator Loss: 1.0631 ___ Generator Loss: 1.1455
Epoch 1/1 ___  Discriminator Loss: 1.1794 ___ Generator Loss: 0.9162
Epoch 1/1 ___  Discriminator Loss: 1.3181 ___ Generator Loss: 0.6624
Epoch 1/1 ___  Discriminator Loss: 1.1037 ___ Generator Loss: 1.2899

Epoch 1/1 ___  Discriminator Loss: 0.4859 ___ Generator Loss: 3.1403
Epoch 1/1 ___  Discriminator Loss: 0.8269 ___ Generator Loss: 2.1203
Epoch 1/1 ___  Discriminator Loss: 1.5807 ___ Generator Loss: 0.4288
Epoch 1/1 ___  Discriminator Loss: 1.5687 ___ Generator Loss: 2.9351
Epoch 1/1 ___  Discriminator Loss: 1.4889 ___ Generator Loss: 0.5515
Epoch 1/1 ___  Discriminator Loss: 0.8764 ___ Generator Loss: 1.6197
Epoch 1/1 ___  Discriminator Loss: 0.5760 ___ Generator Loss: 2.2908
Epoch 1/1 ___  Discriminator Loss: 1.2456 ___ Generator Loss: 0.7994
Epoch 1/1 ___  Discriminator Loss: 1.1491 ___ Generator Loss: 1.0813
Epoch 1/1 ___  Discriminator Loss: 0.9876 ___ Generator Loss: 1.2348

Epoch 1/1 ___  Discriminator Loss: 1.0892 ___ Generator Loss: 0.9508
Epoch 1/1 ___  Discriminator Loss: 0.6396 ___ Generator Loss: 1.9373
Epoch 1/1 ___  Discriminator Loss: 0.8901 ___ Generator Loss: 1.9712
Epoch 1/1 ___  Discriminator Loss: 0.7296 ___ Generator Loss: 1.5962
Epoch 1/1 ___  Discriminator Loss: 1.0283 ___ Generator Loss: 0.9724
Epoch 1/1 ___  Discriminator Loss: 1.7326 ___ Generator Loss: 0.4183
Epoch 1/1 ___  Discriminator Loss: 1.0224 ___ Generator Loss: 1.0930
Epoch 1/1 ___  Discriminator Loss: 1.2312 ___ Generator Loss: 0.6928
Epoch 1/1 ___  Discriminator Loss: 0.9862 ___ Generator Loss: 1.0521
Epoch 1/1 ___  Discriminator Loss: 1.0259 ___ Generator Loss: 1.9112

Epoch 1/1 ___  Discriminator Loss: 0.8160 ___ Generator Loss: 1.4728
Epoch 1/1 ___  Discriminator Loss: 0.7857 ___ Generator Loss: 1.5148
Epoch 1/1 ___  Discriminator Loss: 1.0241 ___ Generator Loss: 1.0514
Epoch 1/1 ___  Discriminator Loss: 1.0514 ___ Generator Loss: 1.0806
Epoch 1/1 ___  Discriminator Loss: 1.1291 ___ Generator Loss: 0.7983
Epoch 1/1 ___  Discriminator Loss: 0.7102 ___ Generator Loss: 2.4975
Epoch 1/1 ___  Discriminator Loss: 1.2057 ___ Generator Loss: 2.0969
Epoch 1/1 ___  Discriminator Loss: 1.2983 ___ Generator Loss: 0.5753
Epoch 1/1 ___  Discriminator Loss: 1.0601 ___ Generator Loss: 0.9046
Epoch 1/1 ___  Discriminator Loss: 1.1330 ___ Generator Loss: 1.6442

Epoch 1/1 ___  Discriminator Loss: 1.0626 ___ Generator Loss: 0.8978
Epoch 1/1 ___  Discriminator Loss: 1.7846 ___ Generator Loss: 2.9055
Epoch 1/1 ___  Discriminator Loss: 1.4364 ___ Generator Loss: 0.6069
Epoch 1/1 ___  Discriminator Loss: 0.7885 ___ Generator Loss: 2.9396
Epoch 1/1 ___  Discriminator Loss: 1.1544 ___ Generator Loss: 0.8468
Epoch 1/1 ___  Discriminator Loss: 1.2763 ___ Generator Loss: 0.6860
Epoch 1/1 ___  Discriminator Loss: 1.3276 ___ Generator Loss: 0.6291
Epoch 1/1 ___  Discriminator Loss: 0.9714 ___ Generator Loss: 2.4828
Epoch 1/1 ___  Discriminator Loss: 1.3277 ___ Generator Loss: 2.7646
Epoch 1/1 ___  Discriminator Loss: 1.1138 ___ Generator Loss: 1.0311

Epoch 1/1 ___  Discriminator Loss: 1.1455 ___ Generator Loss: 0.8125
Epoch 1/1 ___  Discriminator Loss: 1.1208 ___ Generator Loss: 0.8699
Epoch 1/1 ___  Discriminator Loss: 0.9322 ___ Generator Loss: 1.6763
Epoch 1/1 ___  Discriminator Loss: 1.0201 ___ Generator Loss: 1.3174
Epoch 1/1 ___  Discriminator Loss: 0.9608 ___ Generator Loss: 1.3762
Epoch 1/1 ___  Discriminator Loss: 0.9754 ___ Generator Loss: 1.1832
Epoch 1/1 ___  Discriminator Loss: 1.1822 ___ Generator Loss: 0.7892
Epoch 1/1 ___  Discriminator Loss: 1.1908 ___ Generator Loss: 0.8882
Epoch 1/1 ___  Discriminator Loss: 1.4910 ___ Generator Loss: 3.6936
Epoch 1/1 ___  Discriminator Loss: 1.4004 ___ Generator Loss: 0.6216

Epoch 1/1 ___  Discriminator Loss: 1.2057 ___ Generator Loss: 0.7422
Epoch 1/1 ___  Discriminator Loss: 1.1911 ___ Generator Loss: 0.9150
Epoch 1/1 ___  Discriminator Loss: 1.0906 ___ Generator Loss: 1.0518
Epoch 1/1 ___  Discriminator Loss: 1.0649 ___ Generator Loss: 0.8812
Epoch 1/1 ___  Discriminator Loss: 1.3634 ___ Generator Loss: 0.5985
Epoch 1/1 ___  Discriminator Loss: 1.1703 ___ Generator Loss: 0.8106
Epoch 1/1 ___  Discriminator Loss: 1.4856 ___ Generator Loss: 1.9546
Epoch 1/1 ___  Discriminator Loss: 1.1320 ___ Generator Loss: 1.1483
Epoch 1/1 ___  Discriminator Loss: 0.9696 ___ Generator Loss: 1.0945
Epoch 1/1 ___  Discriminator Loss: 1.1271 ___ Generator Loss: 0.8059

Submitting This Project¶

When submitting this project, make sure to run all the cells before saving the notebook. Save the notebook file as "dlnd_face_generation.ipynb" and save it as a HTML file under "File" -> "Download as". Include the "helper.py" and "problem_unittests.py" files in your submission.